Preliminary Characterization of a NAPL-ContaminatedSite using Borehole Geophysical Techniques

Jonathan B. Ajo-Franklin Stanford University, Stanford, CA.Jil T. Geller Lawrence Berkeley National Laboratory, Berkeley, CA.

Ernest L. Majer Lawrence Berkeley National Laboratory, Berkeley, CA.John E. Peterson Lawrence Berkeley National Laboratory, Berkeley, CA.

Kenneth H. Williams Lawrence Berkeley National Laboratory, Berkeley, CA.Jerry M. Harris Stanford University, Stanford, CA.

Abstract

We present preliminary results from an on-going geophysical investigation of the former DOE Pinel-las site, a site with confirmed non-aqueous phase liquid (NAPL) contamination. Our eventual goalis the effective use of integrated crosswell geophysical methods, specifically seismic and radar tomog-raphy, to remotely detect zones of high NAPL saturation. In this paper we discuss our motivationand early efforts at using a combination of seismic and radar information to understand site ge-ology and possible contaminant signatures. We also present our current approach to consistentacquisition, preprocessing, tomographic inversion, and interpretation of joint seismic/radar profiles.Although direct geophysical evidence of NAPL pools has not yet been obtained, several regions ofanomalous seismic attenuation were detected. These attenuating regions could not be explainedby borehole effects or lithology and may be caused by regions of partial gas or NAPL saturation.Possible gas sources include biogenic production from contaminated regions and side-effects fromprevious remediation activities. Continuing research at the Pinellas site will focus on constrainingthe mechanism responsible for the observed seismic attenuation, developing a more complete modelof site soil properties, and applying more quantitative approaches to the integration and analysisof our seismic and radar images.

1 Introduction: The Pinellas Project

The detection and delineation of non-aqueous phase liquids (NAPLs) using geophysical methodsis a challenging problem which has been considered, with varying degrees of success, for bothcontrolled release scenarios [7] [6] [29] [27] and previously existing contaminant sites [31] [13]. Anoutstanding problem with previous field studies is the difficulty in distinguishing the geophysicalsignatures of NAPLs from existing lithological variations which cause similar changes in physicallyobservable properties, such as dielectric constant. This problem is particularly striking when onlysurface GPR data is available for imaging purposes.

1

Driven by recent core scale and physical modeling studies of the acoustic properties of NAPLs[11] [30] [16] [10], we have embarked on a pilot project to geophysically detect NAPLs using acombination of crosswell radar and seismic methods with further constraining information derivedfrom cone penetrometry and natural gamma measurements. This paper presents preliminary resultsfrom our on-going geophysical investigation of a former U.S. Department of Energy (DOE) facilityin Pinellas County, Florida, a site with confirmed NAPL contamination. In addition to detailingthe site, acquisition strategy, and processing approach we will consider the general problem ofintegrating seismic and radar tomograms for environmental characterization purposes. Two regionsof the site, one within the zone of known NAPL contamination (the “dirty” zone) and one outsidethe suspected NAPL zone but still within the dissolved contaminant plume (the “clean” zone) willbe examined in detail.

Although direct geophysical evidence of NAPL pools has not yet been obtained, several regionsof anomalous seismic attenuation were detected. These attenuating regions could not be easilyexplained by borehole effects or lithology and may be caused by regions of partial gas or NAPLsaturation. Possible gas sources include biogenic production from contaminated regions and side-effects from previous remediation activities.

1.1 Site History, Geology, and Existing Evidence for NAPLs

The site of our ongoing investigation is a former DOE manufacturing facility located in PinellasCounty, Florida, between the cities of Clearwater and St. Petersburg. From 1956 to 1994 the Pinel-las DOE Plant fabricated neutron generators, thermal batteries, vacuum switch tubes, and a varietyof secondary electronic and mechanical components required for nuclear weapon construction. Overthe lifetime of the facility, several chlorinated solvents were used in degreasing and cleaning ca-pacities. Between 1957 and 1968 a small region in the Northeast section of the site, immediatelywest of the East Pond, was used as a drum disposal and staging area. Excavation at the North-east Site in the latter half of 1994 revealed construction debris and drums with solvent and resinresidues. The contaminants discovered in groundwater and soil sampling included trichloroethene,methylene chloride, dichloroethane, vinyl chloride, benzene, toluene, oil, and various unidentifiedresin compounds. Free-phase volatile organic compounds (VOCs) were also extracted from severalmonitoring and recovery wells, indicating the presence of a separate contaminant phase in someregions of the site. Additional evidence for NAPLs at the Northeast site was obtained during a1996/1997 field demonstration of a dual auger rotary steam-stripping system in which high levelsof VOCs overloaded the catalytic treatment system. Of the wells visible on Figure [1], free phaseVOCs have been extracted from both RW03 and RW06.

The geology of our section of the NE site, as determined from a large number of visuallycategorized soil borings and cone penetrometry (CPT) push-points, consists of a relatively one-dimensional sequence with more subtle lateral variations. A shallow surficial aquifer consisting offine sands and silty sands extends from the surface to a depth of approximately 10 − 11m. Thesurficial aquifer gradually terminates against the Hawthorn Group, an 18 − 21m thick region ofsilty-clays, silty-sandy clays, and carbonaceous clays. Some finer structures including zones of shellfragments, phosphate nodules,and increased silt content exist in the surficial aquifer. A definite andlaterally continuous textural transition visible in all of the recorded natural gamma logs suggests

2

ROI 1

ROI 2

Pinellas NE Site : LBNL Study Region

*

CP25

CP09

C P 52

CP23

CP08CP07

CP0

R

5

W12

M 34D

M31D

533−15

G5

G4

G3

G2

G1SB37

SB38

SB34

SB40M17D

SB39

M17S

G

G

2

G

1

0

1

8

9

G17G16

G15

4

G13

G12

SB65

S109

SB43

G1

RW 06

RW 03

SB5

*

SB92

SB16

SB

S121SB17

58

Plume TransitionRegion

Plume TransitionRegion

CPT Push Point

Soil Boring (SB) or Monitoring Well (M)

Cased CPT Hole (2in)

Survey Well (4in)

Region Of Interest (ROI)

Crosswell Seismic Survey Profiles

Crosswell Radar Survey Profiles

Plume Transition BoundaryWell With Recovered NAPL*Well Where NAPLs Visually Noted*

*

Clean Zone

TCE/Toluene Plume

Clean Zone

TCE/Toluene Plume

0 10 20 30 405Feet

Figure 1: Pinellas NE Site Study Area

an increase in mineralogical clay content near an absolute elevation 1 of −0.5m.

1.2 Acquired Datasets

The primary data sets acquired for our geophysical characterization of the of Pinellas NE site (seeFigure [1]) consisted of a curtain of crosswell seismic profiles extending from a zone known to haveNAPL contamination [ROI 1] to a region where aqueous phase contaminant levels were an order ofmagnitude lower but still above safe limits [ROI 2]. The crosswell seismic datasets were acquiredusing a 24 channel Geometrics Geode acquisition system, a 24 channel hydrophone string, and afluid coupled piezoelectric source. Nineteen seismic sections were acquired at interwell spacingsranging from 2 to 8 meters. For most crosswell seismic profiles, usable signal at up to 6.5 khz wasrecorded.

Crosswell radar datasets were acquired using a Sensors & Software Pulse-Ekko 100 BoreholeSystem and 200 mhz antennas. Radar data could not be acquired in all of the wells used for seismicmeasurements due to the presence of conductive (steel) casing at some locations. An attempt wasmade to match geometries for the seismic and radar surveys by using the same source/receiverspacing (1/8 m). In addition to the crosswell measurements, a Mount Sopris MGX II system wasused to acquire gamma and conductivity logs in all of the relevant wells for use in constraininglithology. Several existing CPT push-points were also available for lithology comparisons althoughthe lack of site-specific CPT calibration data made lithology designations less certain.

1All elevation references are made with respect to sea-level in terms of Florida State Plane coordinates. Surfaceelevation is approximately 6m above sea-level.

3

IncreasingS/N

Low

erSu

rfic

ial

Aqu

ifer

P−wave Velocity (m/s)

ROI (2)

G15 G14 G13 G12 G16 G17 M34DG18 RW12 G19 G20 M31D

Water Table

Ele

vati

on (

m)

Upp

er S

urfi

cial

Aqu

ifer

Preliminary Site−Wide Vp TomographyInside NAPL Area Zone Of Uncertainty Outside Of NAPL Area

1600 1700 1800

Attenuating Region

ROI (1)

Figure 2: Preliminary Site-wide Vp Imaging Results

1.3 Preliminary Site-Wide Tomography

Two separate tomographic processing flows were applied to the crosswell data. The first flow wasaimed at getting an initial set of site-wide tomographic images for immediate interpretation andthe second focused on techniques for combined interpretation of the seismic and radar datasets.The initial traveltime tomography used the preprocessing steps and reconstruction techniques de-scribed in references [24] and [25] and essentially consists of 2D straight ray tomography with adamped Algebraic Reconstruction Technique (ART) solver. Although we will not describe the pre-liminary processing flow in detail, the referenced approach has been applied successfully to a varietyof environmental imaging problems. The second tomographic flow was tailored to the combinedinterpretation problem and will be described in detail in the next sections.

Figure [2] depicts the preliminary site-wide P-wave velocity (Vp) section derived from nine differ-ent crosswell seismic profiles. The section stretches from well G15 (see Figure [1]), which is locatedwithin the general region of NAPL contamination, to well M31D located in a clean zone with loweraqueous phase contaminant levels. A noticeable feature is a laterally extensive zone of low signalstrength which made picking difficult, when possible at all. In the preliminary processing run, pixelswith insufficient ray coverage due to picking difficulties were eliminated from the inversion. Thezone of high attenuation gradually decreases in expanse when moving towards the clean zone andis not visible in the last two crosswell profiles. The attenuating region is not consistent with anylithologic trends observed in well logs and may be the result of biogenic gas production associatedwith highly contaminated regions of the site. A more detailed discussion of the observed attenuationand various possible mechanisms is continued in Section 5.

2 Integrated Geophysics at Pinellas

The term “Integrated Geophysics” has several different meanings within the diverse community

4

of geoscientists. On the most basic level, integrating geophysical data is simply drawing a conclusionfrom multiple sets of available measurements. On the other extreme, integration refers to a completequantitative fusion of all available data to yield a comprehensive earth model with describableresolution and error characteristics. Relatively few previous studies have attempted to integrateradar and seismic measurements and almost all have attempted integration on a purely interpretivebasis after separate collection and processing of both datasets [5] [4]. Although we do not yet havea suitable strategy for a truly consistent joint inversion of our crosswell data, we have attemptedto tailor our acquisition and processing flow to allow comparison and combined interpretation ofthe two datasets. In this section we will review the petrophysical motivation for data integration interms of NAPL detection, an acquisition and preprocessing strategy designed to make our seismicand radar datasets comparable, our current tomographic imaging flow, and some possible routesfor quantitative integration of tomography results.

2.1 A Petrophysical Motivation

One of the goals of the Pinellas project is the quantitative integration of crosswell seismic andcrosswell radar data for use in characterizing subsurface fluid properties, specifically detecting re-gions of high NAPL saturation. NAPLs, including trichloroethene (TCE) ,the mostly likely NAPLin our study region, are generally non-polar with low dielectric constants (3.3 for TCE [22]) incomparison to water which has a dielectric constant of approximately 80 [1]. This property makesground penetrating radar a good candidate technology for NAPL detection. The extensive Bordenfield tests [7] [6] [29] [27] convincingly demonstrated that high levels of NAPL are visible usingGPR in a time-lapse sense if prior knowledge of the spill location is available. However, detectionof NAPLs using dielectric properties alone seems difficult since similar radar signatures could begenerated by lithological changes, including variations in matrix porosity or clay content. Recentwork on the acoustic properties of NAPL saturated unconsolidated sediments [11] [30] [10] opens upthe possibility of using seismic velocity or attenuation data in tandem with dielectric measurementsderived from radar to more reliably predict the presence of NAPLs.

Figure [3] depicts a theoretical cross-plot of Vp and dielectric constant (κ) for a clean quartzsand subject to two property trends, one scenario where porosity is varied from 0.2 to 0.45 with azero TCE saturation (100% water saturated , shown in red) and a second where porosity is fixedat 0.36 and TCE saturation is varied from 0 to 80% (shown in blue). The complex refractive indexmodel (CRIM) was used to estimate effective dielectric properties. Vp was estimated by using Hertz-Mindlin contact theory [21] to determine the dry frame modulus and then saturating the pack usingGassmann’s equation [9] [17] with the properties of the TCE/water mixture calculated using theReuss effective fluid model [28] [17]. If only dielectric constant is considered, there is no reasonableway to determine whether the source of an observed radar anomaly is due to a variation in porosityor the existence of NAPLs. While we do not expect this particular combination of effective mediumtheories to be valid at most sites, the important insight gained from Figure [3] is the possibility ofusing a second data attribute, in this case Vp, to unravel lithologic trends from NAPL saturationtrends.

5

5 10 15 20 25 301500

1600

1700

1800

1900

2000

2100

2200

Effective Dielectric Constant

V p (

m/s

)

Theoretical Vp/Dielectric Crossplot : DNAPL vs. Porosity Trends

Phi=0.36,TCE=0

Phi=0.36,TCE=0.8

Phi=0.2,TCE=0

Phi=0.45,TCE=0

Trend : Increasing Porosity

Trend : Increasing TCE Saturation

Figure 3: Theoretical Vp/dielectric constant cross-plot for a sand with either TCE saturation orporosity variations (phi = porosity)

6

2.2 Quantitative Integration of Tomography Results: Pitfalls

While cross-plots like Figure [3] are commonly generated from laboratory core measurements orhigh quality logging data, they are seldom used for analyzing tomography results due to a host ofinconsistencies in the integration step. Ultimately, we would like to apply appropriately upscaledpetrophysical relationships to a set of tomographic images with different data sources (dielectricconstant, P and S-wave velocity etc.) and derive either secondary properties such as fluid saturationcharacteristics or more heuristic attributes which might provide a useful proxy indicator for NAPLs.

One of the most significant barriers to the quantitative use of even individual tomographic imagesis the presence of artifacts in the final inversion result, possibly caused by aperture limitation, irreg-ular survey coverage, errors in picking, positioning, or discretization, and even fundamental physicalinconsistencies between the way wave propagation actually occurs and the modeling scheme embed-ded in the tomography algorithm. At the integration step we are also forced to consider inevitablequestions of scale; seldom are the resolving powers of two different geophysical measurements identi-cal. This implies that even after inversion, seismic and radar images have different spatial averagingproperties.

2.3 Preprocessing

After data collection we performed two stages of preprocessing. The first stage involved verystandard operations, specifically file format conversion, the addition of trace geometry information,the manual removal of very low S/N traces, low cut filtering, and traveltime picking. One of thelargest differences observed between the radar and seismic datasets was the mismatch of angularapertures; while high angle rays in the seismic profiles could often be picked, the radar profileswere typically limited to an angular aperture of approximately ±45 degrees from the horizontal dueto the radiation pattern of the source and receiver antennas. The second stage of preprocessingaimed to make the data footprints of the seismic and radar surveys more similar in order to reduceartifacts when the separately processed tomograms were combined. The angular aperture of theseismic data was truncated on a source-by-source basis to conform with maximum aperture presentin the radar dataset. In regions where both seismic and radar picks were available, we had sufficientspatial sampling to interpolate both datasets onto the same source/receiver position grid to exactlymatch portions of the survey geometry. In zones where source gathers from one of the two datasetswere not pickable at low angle, no matching step outside of ±45 degree aperture reduction wasapplied. This resulted in some coverage inconsistencies, particularly in the high seismic attenuationzones visible in Figure [2].

2.4 Traveltime Tomography Techniques

Our goal in designing the second tomographic processing flow was to choose an algorithm whichcould consistently recover from poor data coverage in the high attenuation zone, provide imagesrelatively free of ray-based artifacts, and would require a minimal number of user controlled “tuning”parameters which could bias the tomography results. While we are ultimately moving towardsapproaches which truly honor the physical resolving power of band-limited data by incorporatingfresnel regions [20] [14], our current implementation only accomplishes this indirectly through useof convolutional quelling [19].

7

All ray-tracing was performed using straight rays, a reasonable assumption given the small(generally < 10%) velocity contrasts present within the primary imaging zone. One region wheresignificant GPR ray-bending occurred was within the vicinity of the capillary fringe. However, sincethe vadose transition region was not one of our imaging targets, the few rays which crossed thiszone were discarded.

The tomographic imaging technique we decided to use is a two-step process which first generatesa 1D model and then uses this layered image as a prior for a 2D inversion, similar to the techniquedescribed in [35]. The 1D problem is usually very well resolved. Even in regions where significantgaps in data coverage exist, there are usually enough rays passing through a given layer to allow avelocity estimate, although the variance of the resulting estimate may be high. The layered inversionis parametrized as a stack of equal height, constant slowness sections. Regularization constraintsare added with a one-dimensional 1st derivative operator which favors flat models. The generalizedinverse for the resulting system, with the addition of the regularization term is

G−g1 = [GT

1 G1 + λ21 (W T

m1Wm1) ]−1 GT1 (1)

where G1 is the kernel for the layered inversion problem, λ1 is a regularization parameter, andWm1 is the 1st derivative model weighting matrix. Wm1 uses forward differences with appropriatetreatment of boundary points. The resulting 1D model estimate can then be written in terms ofthe generalized inverse as

m1est = G−g1 d . (2)

Since the layered model has only a small number of parameters, the generalized inverse is calculatedusing the singular value decomposition (SVD).

The Generalized Cross-Validation technique (GCV) [12] [33] is used to determine an appropriateregularization parameter for the 1D problem without having to assume a-priori a particular noiselevel. The GCV technique attempts to choose a regularization parameter resulting in a “robust”solution, i.e. a λ value which allows reliable prediction of the model if any single data point isdiscarded. GCV C(λ), the metric which the GCV technique uses to judge the fitness of a particularchoice of regularization or parametrization is

GCV C(λ) =‖(I − H1) d‖2

Trace(I − H1)2(3)

where H1, the data resolution matrix, implicitly contains λ1 since H1 = G1 G−g1 . The GCV criterion

is minimized over the space of reasonable regularization parameters. This is a 1D minimizationproblem requiring calculation of G−g

1 and H for a large number of λ1 values. The λ1 value thatminimizes equation [3] is then used to evaluate the 1D tomography solution [2]. One commonproblem with the GCV approach is that GCV metric (3) often has a wide shallow minimum resultingin uncertainty as to the appropriate λ [15] although we did not have this difficulty in our 1Dinversions.

The 2D tomography problem is parametrized as a rectilinear mesh of constant slowness pixelsand is solved using m1est as a prior. In reality, m1est is not really prior information since it is derivedfrom the same data uesed to estimate m2est. However, including a layered prior for the 2D problemgenerates a solution which effectively “defaults” to a 1D model in regions where the 2D solution is

8

poorly resolved. If we let G2 be the kernel for the 2D tomography problem, the solution with our1D prior and a natural generalized inverse would be

m2est = m1est + G−g2 [d−G2 m1est] . (4)

A bit of algebraic manipulation and use of the definition of the model resolution matrix, R = G−gG,yields the alternative form,

m2est = G−g2 d + (I −R2) m1est (5)

which shows that information from the 1D solution is only used in cases where the 2D solutionlacks sufficient resolution. If the 2D model is perfectly resolved (R2 = I ), the resulting tomogramdoes not include any information from the layered estimate. Since computation of R2 is expensive,equation [4] is solved unless the model resolution matrix is desired for more complete analysis ofthe solution. To maintain dimensional consistency, the layered prior used in the 2D problem isresampled onto the 2D grid; typically the same depth spacing is used for both inversions so nointerpolation step is required. The end result of the 1D/2D tomography process are images whichprovide reasonable 1D velocity estimates in regions with poor ray coverage without sacrificing lateralresolution in other sections of the profile.

While the kernel for the 2D problem, G2, is initially computed using infinitesimal rays, theserays are broadened using the convolutional quelling technique [19] [26] to heuristically account forthe effects of band-limitation in both our radar and seismic experiments. The maximimum widthof the quelling filter is chosen to be a 1

2wavelength, where a single wavelength value derived from

the mean velocity of the 1D model and the median centroid frequency of all incorporated traces isused for the entire kernel. The resulting fat rays limit the resolution of the final image to levelsconsistent with the recorded signal bandwidth. However, these broadened rays are not the idealapproximation to fresnel zones since they lack variations in width along the length of the ray andthe appropriate off-axis weighting properties. In this paper we do not explicitly show the kerneland model reweighting steps from the convolutional quelling technique in our derivations. Reference[19] provides complete details on the implementation of convolutional quelling in a least-squarescontext.

The natural generalized inverse (G−g2 ) for the 2D problem (5) is computed using the truncated

singular value decomposition (TSVD) [18]. If we write the SVD of G2 as

G2 = U2 Λ2 V T2 (6)

where U2 and V2 are the left and right eigenvectors and Λ is the diagonal eigenvalue matrix, thenwe can write the natural generalized inverse of G2 as

G−g2 = V2p Λ−1

2pUT

2p(7)

where the p subscript denotes truncation at the pth eigenvalue. Initially, we attempted to use theGCV criterion to compute an appropriate cut-off point for the SVD truncation but irregularities inthe GCVC curve made this estimation difficult. The presence of correlated noise components in thetraveltime pick data may be partially responsible for the failure of the GCV technique for the 2Dinversion [15]. For larger problems, iterative techniques such as LSQR [23] [32] coupled with theL-Curve method of regularization parameter determination would be more appropriate than use of

9

the SVD + GCV flow due to the SVD’s asymptotic computational cost of O(m n2) for m data andn model parameters and the fact that computation of the GCV metric requires the data resolutionmatrix (H).

We typically calculate both the seismic and radar tomograms on matching rectilinear grids sothat both dielectric constant and Vp values are available at the same spatial locations. However,this superposition of meshes does not mean that the averaging properties of the radar and seismickernel are identical since the quelling widths and regularization parameters for each dataset aredetermined independently.

2.5 Post-Processing and Rock Physics Integration

With radar and seismic tomograms calculated for the same profile, the final integration of the twodatasets for NAPL detection can take several forms. In the simplest case where only qualitativeinterpretation is performed on the two datasets, we line up our tomograms with any axillary infor-mation including core descriptions, well logs, and water sampling data and look for regions withinthe saturated zone with low dielectric constant values, low seismic velocities, and no consistentlithological explanation. To interpret more subtle variations in properties, a quantitative approachbased on rock physics would be desirable. Ideally, an inverse petrophysical model which consistentlydescribes both the dielectric and acoustic properties of sediments partially saturated with NAPLscould be used to combine the two tomograms to yield a map of porosity and NAPL saturation. Inpractice this is generally not feasible due to the physical complexity of unconsolidated sediments,the presence of variations in clay content, and a variety of scaling errors in both petrophysicalrelationships and the imaging process. A slightly less ambitious approach might involve the use ofproperty cross-plots generated from tomographic images to identify trends associated with NAPLsaturation. Although the cross-plot trends might not be equivalent to those generated from theo-retical or core-scale experimental models, general features could provide guidance in the qualitativeseparation of litholgy from saturation effects. Although we have generated Vp/dielectric cross-plotsfor several well pairs, we are currently carrying out laboratory calibration measurements to betterunderstand the observed proprerty trends.

3 Two Integrated Seismic/Radar Crosswell ProfilesWe now apply the processing approach described in Section 2 to two crosswell profiles at the

Pinellas NE site, one within the NAPL contaminated region (Figure [4], ROI 1, G14-G15) and onewithin a cleaner portion of the site (Figure [5] ROI 2 : G19-G20). Both Figures show local CPTlithology estimates and natural gamma logs in addition to crosswell radar and seismic tomograms.Figure [6] shows the tomograms plotted with the same color scale for property comparisons betweenthe two well pairs. Figures [5] and [4] use different linear colorscales to enhance smaller anomalies.Due to the high seismic attenuation and consequently poor ray coverage, the region between -1 and-3 m elevation in the G14-G15 seismic profile (Figure [4]) only includes 1D model variations; thisis a result of our hybrid 1D/2D formulation (see equations [4] [5]) . This region, labeled (Z1) inFigures [4] and [6], also has the largest velocity uncertainty since the weak high angle rays whichpass through this zone are the most difficult to pick.

The top of the Lower Surficial Aquifer was consistently located in the crosswell radar tomogramsand matched the CPT derived textural transition to within 20 cm. The radar surveys also detected

10

a subtle textural transition at an elevation of -0.25 m, a region with a gradual increase in fine particlecomponents which is also visible on the gamma logs. The observed transitions in radar velocity canbe largely explained by changes in volumetric water content due to variations in porosity/sorting, orin the case of the lower aquifer boundary, water trapped within the clay component. More strikingthan the features visible within the crosswell radar images is the lack of a high radar velocity regionnear the Hawthorn formation or the Lower Surficial Aquifer boundary in the G14-G15 profile, whichwe expected to be the best signature of a zone of high NAPL saturation. In fact, the absolute radarvelocities were higher for the lower surficial aquifer in the G19-G20 profile, suggesting that theNAPL saturations in the ROI 1 contaminated region are low, if present.

200 400−4

−3

−2

−1

0

1

2

3

4

CPS

G03:GAM

F−Ratio (%)

G04:CPT

0 5 10−4

−3

−2

−1

0

1

2

3

4

1400

1450

1500

1550

1600

1650

1700

1750

1800

1850

1900

Lateral Position (m) Vp (m/s)

G14−G15 Vp Tomogram

0 1 2−4

−3

−2

−1

0

1

2

3

4

5.6

5.8

6

6.2

6.4

6.6

6.8

V_em(x10^7 m/s)

Lateral Position (m)

Ele

vatio

n (m

)

G15−G14 Radar Velocity Tomogram

0 1 2−4

−3

−2

−1

0

1

2

3

4

Clay

Sand

CPT Lithology

(Dirty Zone)

ROI 1

G14−G15Gamma TexturalTransition

VadoseTransition

Hawthorn Formation

Upper

(Aquitard)

LowerSurficialAquifer

UpperSurficialAquifer

(Z1)

Figure 4: ROI 1 : G14-G15 : Dirty Well Pair

3.1 Working towards Quantitative Integration

Despite questions of scale and resolution consistency between the seismic and radar tomograms,we have started to examine Vp/dielectric crossplots generated from the Pinellas tomograms tohelp understand the trends associated with natural lithological transitions. Figure [7] depicts theVp/dielectric crossplot for every point in both the G19-G20 and the G14-G15 profiles. The internalstructure of the crossplot is dominated by vertical property variations with smaller lateral variationsgenerating the internal “loops” in both curves. The region in Figure [7] labeled (Z1) corresponds tocross-plot points within the poorly resolved lower region of the G14-G15 seismic profile and shouldbe treated with some caution. We have not yet made the core-scale calibration measurementsrequired to interpret Figure [7] with confidence but we are currently in the process of building a dualdielectric/acoustic property model based on samples extracted from several locations at the PinellasNE site. Using ultrasonic pulse transmission techniques and time domain reflectometry (TDR) wehope to both unravel some of the cross-plot’s structure and make some more quantitative statementsconcerning the sensitivity of radar measurements to NAPL saturation within the different lithologiespresent at Pinellas.

11

200 400−4

−3

−2

−1

0

1

2

3

4

CPS

RW12:GAM

F−Ratio (%)

CP09:CPT

0 5 10−4

−3

−2

−1

0

1

2

3

4

Ele

vatio

n (m

)

1600

1650

1700

1750

1800

Vp (m/s)Lateral Position (m)

G19−G20 Vp Tomogram

0 1 2−4

−3

−2

−1

0

1

2

3

4

6.2

6.4

6.6

6.8

7

7.2

Lateral Position (m) V_em

(x10^7 m/s)

G19−G20 Radar Velocity Tomogram

0 1 2−4

−3

−2

−1

0

1

2

3

4

Clay

Sand

CPT Lithology

(Clean Zone)

G19−G20

ROI 2

Gamma TexturalTransition

VadoseTransition

LowerSurficialAquifer

Hawthorn Formation

Upper

(Aquitard)

UpperSurficialAquifer

Water Table

Figure 5: ROI 2 : G19-G20 : Clean Well Pair

Lateral Position (m)

Ele

vatio

n (m

)

Ele

vatio

n (m

)

G14−G15 (ROI 1) S

0 1 2

−3

−2

−1

0

1

2

3

Lateral Position (m)

G19−G20 (ROI 2) S

0 1 2

−3

−2

−1

0

1

2

3

Lateral Position (m)

G14−G15 (ROI 1) R

0 1 2

0

−1

−2

−3

1

2

3

Lateral Position (m)

G19−G20 (ROI 2) R

0 1 2

−3

−2

−1

0

1

2

3

1450

1500

1550

1600

1650

1700

1750

1800

1850

1900

5.6

5.8

6

6.2

6.4

6.6

6.8

7

(Z1)Vp (m/s)

Seismic Velocity Comparison

Vem (m/s)*107

Radar Velocity Comparison

Figure 6: ROI 1 + ROI 2 Tomographic Comparison : Same Color Scale

12

18 20 22 24 26 28 301400

1500

1600

1700

1800

1900

2000

(Z1)

G14−G15 DataG19−G20 Data

Cro

ssw

ell S

eism

ic V

p E

stim

ate

(m/s

)

Crosswell Radar Dielectric Constant Estimate

Seismic/Radar Crossplot (G19−G20 + G14−G15)

Figure 7: Experimental seismic/radar crossplot for the G19-G20 and G14-G15 well pairs derivedfrom crosswell tomography measurements

13

Another problem with the integration of crosswell seismic and radar measurements at Pinellasis the large difference in resolution between the two measurement techniques. We calculated mapsof the local wavelength of both the seismic and radar profiles for the G19-G20 well pair (Figure [8])using the velocity tomograms and the mean centroid frequency of each survey. The third panel inFigure [8] shows the point-by-point ratio of local seismic wavelength to local radar wavelength, whichranges between 0.42 and 0.55 for the G19-G20 surveys. To truly integrate the seismic and radarimages we must eventually address the different averaging patterns of the two property estimatesby either a wave-consistent upscaling or statistical downscaling approach.

0.31

0.315

0.32

0.325

0.33

0.335

0.34

0.345

0.35

0.355

Lateral Position (m)

Ele

vatio

n (m

)

Seismic Wavelength Est.

0 0.5 1 1.5 2

−3

−2

−1

0

1

2

0.65

0.7

0.75

Lateral Position (m)

Ele

vatio

n (m

)Radar Wavelength Est.

0 0.5 1 1.5 2

−3

−2

−1

0

1

2

0.42

0.44

0.46

0.48

0.5

0.52

0.54

Lateral Position (m)

Ele

vatio

n (m

)

Seismic/Radar Wavelength

Mismatch

0 0.5 1 1.5 2

−3

−2

−1

0

1

2

Wavelength (m)Wavelength (m) Wavelength Ratio

Figure 8: Local estimates of wavelength and wavelength mismatch for the G19-G20 seismic andradar tomograms

4 Understanding the Attenuating Region:

A Gas Mechanism?

A dominant feature in the crosswell seismic datasets collected at Pinellas was the presence of a highlyattenuating region centered around an absolute elevation of aproximately -2 m. Figure [9] shows acomparison of the zero-offset seismic sections for the G19-G20 well pair (ROI 2) and the G14-G15well pair (ROI 1). Label (a) indicates the depth section characterized by very low signal strength,straddling the boundary between the upper and lower surficial aquifers. The zone of attenuation

14

0 01 12 23 34 4

x 10 x 10−3 −3

−4 −4

−3 −3

−2 −2

−1 −1

0 0

1 1

2 2

3 3

4 4

200 400−4

−3

−2

−1

0

1

2

3

4

CPS

G03:GAM

F−Ratio (%)

G04:CPT

0 5 10−4

−3

−2

−1

0

1

2

3

4

Time (s)

G19−G20 Seismic ZOFF

Time (s)

G14−G15 Seismic ZOFF

Ele

vatio

n (m

)

Clay

Sand

CPT Lithology

Gamma TexturalTransition

VadoseTransition

LowerSurficialAquifer

Hawthorn Formation

Upper

(Aquitard)

UpperSurficialAquifer

Water Table

a)

Zero−OffsetSeismicComparison

"Clean" "Dirty"

Figure 9: Comparison of the zero offset sections for the G19-G20 and G14-G15 crosswell seismicprofiles

laterally coincides with the region of the site where high levels of aqueous phase contaminant weremeasured and some direct NAPL samples taken. No corresponding region of unusual dielectricbehavior was noted in this region. When the low signal region was initially observed, we consideredseveral possibilities besides attenuation within the media. A shift in borehole coupling propertieswas ruled out by examination of source and receiver gathers within the attenuating region. All ofthe wells within the ROI 1 and ROI 2 pairs were drilled using rotary sonic and cased with 4 in. PVCpipe. Since velocity contrasts within the crosswell profiles are relatively small, geometric shadowzones also seem unlikeley. Quantitative estimation of attenuation within the low signal zone is adifficult task since few rays with sufficient energy to pick cross the zone.

Lithological, fluid, and gas effects were all initial candidates for the attenuation mechanismbut a purely lithological explanation seemed the least likely due to the lack of any large lateraltextural trends in our section of the NE site. However, until further measurements are made,we cannot rule out lithology as a possible source. Several recent lab-scale investigations haveconsidered seismic attenuation and velocity changes due to the presence of NAPLs [11] [30]. Thesestudies demonstrated that P-wave attenuation is considerably more sensitive to NAPL saturationthan P-wave velocity for lab-scale core measurements. A direct NAPL-patch scattering mechanismmight be a possible source for the attenuating zone. However, using the previously discussedCRIM model as a simple guide, we would expect bulk TCE saturations over 20% to be visibleas an anomalously fast region on the radar tomograms. Additionally, the attenuation mechanismfunctioning in ultrasonic NAPL experiments (100-200 khz) may not be operative at field crosswellfrequencies (1-8 khz). A gas-related mechanism seems to be the most plausible scenario, sincesmall volumetric gas contents can significantly affect seismic wave propogation without a largecorresponding radar signature. On a more detailed level, we can imagine scattering off of gassypatches, a patch equilibration mechanism [8], or direct attenuation related to bubble resonance

15

[2] [3] [34]. However, any conclusive determination of the operative attenuation mechanism wouldrequire either another detailed coring effort with gas and fluid sampling or possibly a secondarysuite of high-resolution crosswell shear wave measurements to constrain the saturation models.

If we assume that a gas mechanism is the primary source of attenuation, several possible gassources could be posited. The high levels of vinyl chloride (VC), a product of the reductive dechlo-rination of TCE or DCE, found in wells M17D and RW06 suggest that contaminant degradationdriven by biogenic activity might be a possible gas source. If this pathway were conclusively proven,gas-generated seismic attenuation might be useful as a proxy indicator for the presence of NAPLsat Pinellas. Non-condensible gas generated by the auger steam-stripping treatment might also beresponsible, although attenuation levels do not seem to be well correllated with proximity to theaugered zones. Pressure reduction, due to groundwater withdrawal for plume containment, and theresultant degassing might also be responsible for the attenuating zone.

5 Conclusion and Future Work

Although our preliminary work at the Pinellas NE site has not yet yielded concrete geophysicalevidence for the presence of NAPLs in the subsurface, we remain hopeful that ongoing improvementsin our image reconstruction and data integration algorithms coupled with a better understandingof site soil properties will allow a more conclusive answer to the remote NAPL detection problem.A detailed ground-truthing process based on careful coring and prompt chemical analysis of theextracted pore fluids would be the best way to test the efficacy of our geophysical analysis. Crosswellshear-wave measurements using locking three-component receivers might also provide us with amore detailed model of saturation-driven seismic anomalies in the near-surface, especially withinthe problematic zone of high attenuation.

Acknowledgments

This work was funded by the Subsurface Contamination Focus Area of the Environmental Man-agement Program of the USDOE, under contract # DE-AC-03-76F0098. We would also like tothank the EPA for funding the first author through the STAR Fellowship program and the Stan-ford Department of Geophysics for funding through the Chair’s Fellowship for Intergroup Research.This research would not have been possible without the cooperation of the DOE Pinellas Site Staffincluding Dave Ingle and the S.M. Stoller Corp. Crew. We would also like to thank Cecil Hoffpauier(LBNL), Philip Rizzo (LBNL), and Jaime Urban (Stanford) for assistance in the collection of fielddata at Pinellas.

References

[1] S.F. Al-Azzawi, A.M. Awwad, A.H. Al-Dujaili, and M̃.K. Al-Noori. Dielectric constants and excessvolumes of 2-Pyrrolidone + Water at several temperatures. Journal Of Chemical Engineering Data,35:463–466, 1990. 5

[2] A.L. Anderson and L.D. Hampton. Acoustics of gas-bearing sediments I. Background. Journal of theAcoustical Society of America, 67(6):1865–1889, June 1980. 16

16

[3] A.L. Anderson and L.D. Hampton. Acoustics of gas-bearing sediments II. Measurments and models.Journal of the Acoustical Society of America, 67(6):1890–1903, June 1980. 16

[4] R. Bachrach and A. Nur. Same wavelength GPR and ultra shallow seismic reflection on a riverpoint bar: Sand stratigraphy and water table complexity. Expanded Abstract, Society Of ExplorationGeophysics Annual Meeting, 1998, 1:840–844, 1998. 5

[5] G.S. Baker, R. Plumb, D.W. Steeples, M. Pavlovic and C. Schmeissner. Coincident GPR and ultra-shallow seismic imaging in the Arkansas River Valley, Great Bend, Kansas. Expanded Abstract,Society Of Exploration Geophysics Annual Meeting, 1998, 1:859–861, 1998. 5

[6] M.L. Brewster and A.P. Annan. Ground-penetrating radar monitoring of a controlled dnapl release:200 mhz radar. Geophysics, 59(8):1211–1221, 1994. 1, 5

[7] M.L. Brewster, A.P. Annan, J.P. Greenhouse, B̃.H. Kueper, G.R. Olhoeft, J.D. Redman, and K̃.A.Sander. Observed migration of a controlled DNAPL release by geophysical methods. Ground Water,33(6):977–987. 1, 5

[8] N.C. Dutta and H. Ode. Attenuation and dispersion of compressional waves in fluid-filled porousrocks with partial gas saturation (white model) - part I: Biot theory, part II: Results. Geophysics,44:1777–1805, 1979. 15

[9] F. Gassmann. Uber die elastizitat poroser medien. Vier. der Natur. Gessellschaft in Zurich, 96:1–23,1951. 5

[10] J.T. Geller, M.B. Kowalsky, P.K. Seifert, and K̃.T. Nihei. Acoustic detection of immiscible liquids insand. Geophysical Research Letters, 27(3):417–420, 2000. 2, 5

[11] J.T. Geller and L.R. Myer. Ultrasonic imaging of organic liquid contaminants in unconsolidatedporous media. Contaminant Hydrology, 19:85–104, 1995. 2, 5, 15

[12] G.H. Golub, M.T. Heath, and G. Wahba. Generalized cross-validation as a method for choosing agood ridge parameter. Technometrics, 21:215–223, 1979. 8

[13] T.W. Griffin and K.W. Watson. A comparison of field techniques for confirming dense nonaqueousphase liquids. Ground Water Monitoring and Remediation, 22(2):48–59, 2002. 1

[14] G.Schuster and A. Quintus-Bosz. Wavepath eikonal traveltime inversion: Theory. Geophysics,58:1314–1323, 1993. 7

[15] P.C. Hansen. Analysis of discrete ill-posed problems by means of the L-curve. SIAM Review,34(4):561–580. 8, 9

[16] M.B. Kowalsky, J.T. Geller, P.K. Seifert, K̃.T. Nihei, and L.R. Myer. Acoustic visibility of immiscibleliquids in poorly consolidated sand. Expanded Abstract, Society Of Exploration Geophysics AnnualMeeting, 1998, page 4, 1998. 2

[17] G. Mavko, T. Mukerji, and J. Dvorkin. The Rock Physics Handbook: Tools For Seismic Analysis InPorous Media. Cambridge University Press, 1998. 5

[18] W. Menke. Geophysical Data Analysis : Discrete Inverse Theory. Academic Press, 1984. 9

17

[19] K.A. Meyerholtz, G.L. Pavlis, and S.S. Szpakowski. Convolutional quelling in seismic tomography.Geophysics, 54(5):570–580. 7, 9

[20] R. Michelena and J. Harris. Tomographic traveltime inversion using natural pixels. Geophysics,56:635–644, 1991. 7

[21] R.D. Mindlin. Compliance of elastic bodies in contact. Journal Of Applied Mechanics, 16:259,268,1949. 5

[22] J. Nath. Ultrasonic velocities, relative permittivities, and refractive indices for binary liquid mixturesof Trichloroethylene with Pyridine amd Quinoline. Fluid Phase Equilibria, 109:39–51, 1995. 5

[23] C.C. Paige and M.A. Saunders. LSQR: An algorithm for sparse linear equations and sparse leastsquares. Trans. Math. Softw., 8:43–71, 1982. 9

[24] J.E. Peterson. The Application of Algebraic Reconstruction Techniques to Geophysical Problems.PhD thesis, 1986. 4

[25] J.E. Peterson. Pre-inversion corrections and analysis of radar tomographic data. Journal Of Envi-ronmental And Engineering Geophysics, 6:1–18, March 2001. 4

[26] W.S. Phillips and M.C. Fehler. Traveltime tomography : A comparison of popular methods. Geo-physics, 56(10):1639–1649, 1991. 9

[27] J.D. Redman. Geophysics and the solvents-in-groundwater program. Symposium on the Applicationof Geophysics to Engineering and Environmental Engineering, Soc. Eng. Min. Expl. Geophys., pages375–382, 1992. 1, 5

[28] A. Reuss. Berechnung der fliessgrenzen von mischkristallen auf grund der plastizitatbedingung fureikkristalle. Zeitschrift fur Angewandte Mathematik und Mechanik, (9):49–58, 1929. 5

[29] K.A. Sander, G.R. Olhoeft, and J.E. Lucius. Surface and borehole radar monitoring of a DNAPLspill in 3d versus frequency, look angle, and time. Symposium on the Application of Geophysics toEngineering and Environmental Engineering, 2:455–469, 1992. 1, 5

[30] P.K. Seifert, J.T. Geller, and L.R. Johnson. Effect of P-wave scattering on velocity and attenuationin unconsolidated sand saturated with immiscible liquids. Geophysics, 63(1):161–170, 1998. 2, 5, 15

[31] T.J. Temples, M.G. Waddell, and W. Domoracki. Non-invasive determination of the location anddistribution of dnapl using advanced seismic reflection techniques. Groundwater, 39(3):465–474,2001. 1

[32] A van der Sluis and H.A. van der Horst. Numerical solution of large, sparse linear algebraic systemsarising from tomographic problems. In G. Nolet, editor, Seismic Tomography With Applications inGlobal Seismology and Exploration Geophysics, chapter 3. 1987. 9

[33] G. Wahba. Spline Models for Observational Data, volume 59 of CBMS-NSF Regional ConferenceSeries in Applied Mathematics. Society For Industrial And Applied Mathematics, 1990. 8

[34] R.H. Wilkens and M.D. Richardson. The influence of gas bubbles on sediment acoustic properties: in situ, laboratory, and theoretical results from Eckernforde Bay, Baltic Sea. Continental ShelfResearch, 18:1859–1892, 1998. 16

18

[35] H.W. Zhou, J.A. McDonald, C.A. Link, and J. Jech. Constraining the magnitude of velocity pertur-bations in traveltime tomography. Journal Of Seismic Exploration, 2:365–380, 1993. 8

19